The vast majority of the empirical literature on income distribution is base on household surveys. More recently, there was a series of papers based on Personal Income Tax (PIT) records and also mixing these two types of data sources. However, Brazil also has a long series of establishment level administrative records seldom used in distributive studies. The best example of these microdata sets is RAIS (Registro Anual de Informações Sociais) source collected by the Labour Ministry with an average of 30 milion observations per year in the last two decades. This note documents the evolution and the main close determinants of earnings inequality in the Brazilian formal sector from 1994 and 2015 using RAIS. First, we show that earnings distribution changes observed in RAIS share some of the trends observed in household surveys. In particular, there is a marked inequality fall occured between 2001 and 2014. Second, schooling explains 31.6% of inequality 2015 level and 26.1% of observed fall between 1994 and 2015. These same statistics for individual firm effects are 57.8% and 80.2%, respectively. Meaning that the gross explanatory power of firm effects to explain inequality in the Brazilian formal labor market surpasses all other variables considered taken together. Thir, RAIS unlike other data sources does not have top coding which permits to measure top earnings. The paper applies J-Divergence analysis to show that in spite of overall inequality fall, the monotonic decrease of earnings growth goes only until the 90 percentile above this point the trend is reverted, especially in the top 0.1%. Mimicking some of the trends found in PITs data. Similarly, in spite of a fall of mean education returns, the share of inequality explained by university graduates also rises 43.7% in the same period.